This invention relates generally to a method of forming a ferroelectric film for use as the dielectric of a capacitor memory element for a nonvolatile memory. More particularly, the present invention relates to a manufacturing method using bismuth layered oxide compounds having the desirable electrical characteristic that is compatible with conventional integrated circuit processing.
It is well known that certain materials such as lead zirconate titanate (PbZr.sub.X Ti.sub.1-X O.sub.3), which is also known as "PZT", are "ferroelectric" in that they can retain a reversible electric polarization in the absence of an applied electric field. Typically, these compounds exist as polycrystalline materials containing grains of perovskite crystalline material.
Other compounds, such as the bismuth layered compounds are also ferroelectric. The structure of bismuth layered compounds consists of layers of perovskite, infinite in two dimensions, separated from each other by a bismuth oxide layer. The perovskite layer can be one, two, three or more perovskite unit cells thick. Each of these thicknesses results in a separate structure type, but has the perovskite layer-bismuth oxide layer alternation, giving rise to the term bismuth "layered" compounds. Examples of ferroelectric bismuth layered compounds include bismuth titanate (Bi.sub.4 Ti.sub.3 O.sub.12)and strontium bismuth tantalate (SrBi.sub.2 Ta.sub.2 O.sub.9), which is also known as "SBT", as well as many others.
The reversible polarization of a ferroelectric material film, for example, must be in a direction perpendicular to the plane of electrodes that transfer the applied electric field to the film so that an appreciable amount of charge can be detected. A stable polarization results from the alignment of internal dipoles with the perovskite crystal units in the ferroelectric material. Application of the electric field across a ferroelectric material causes the alignment of the dipoles in one direction. Reversal of the polarity of the applied field also reverses, or switches, the alignment of the internal dipoles. Application of an electric field exceeding a critical level known as the "coercive voltage", VC, causes dipoles within the ferroelectric material to begin alignment, as is shown in the hysteresis curve 10A of FIG. 1A. An increase in the applied electric field results in a saturation level known as the "saturation voltage", VSAT, wherein substantially all of the dipoles within the ferroelectric material have been polarized in one direction. In FIG. 1A, the coercive voltage is about 1.3 volts and the saturation voltage is about five volts. Note that hysteresis curve 10A has a distended shape, which is indicative of a poorly defined coercive voltage, i.e. many of the dipoles have not yet been switched at the coercive voltage level of applied electric field.
Hysteresis curve 10A therefore demonstrates the electrical behavior of a ferroelectric material such as PZT wherein the x-axis represents the applied electric field and the y-axis represents the resultant polarization or charge. Two stable states 12A and 14A are shown in FIG. 1A, which represent opposite alignments of the internal dipoles after the applied electric field has been removed. The amount of charge released in moving between stable states 12A and 14A is known as the "switched charge." It is also known that the rate of alignment of the dipoles is related to the overdrive voltage, i.e. the amount by which the applied voltage exceeds the coercive voltage. The position of the dipoles and the associated electric charge in response to an applied electric field can be detected with appropriate sensing circuitry. Ferroelectric materials, therefore, can be used as the dielectric material in a ferroelectric capacitor that in turn is used as the memory element in a nonvolatile memory cell.
Hysteresis curve 10B in FIG. 1B demonstrates the electrical behavior of a ferroelectric material such as SBT wherein the x-axis represents the applied electric field and the y-axis represents the resultant polarization or charge. In the hysteresis curve 10B, the coercive voltage is about one volt and the saturation voltage is about 2.5 volts. Note that hysteresis curve 10B has a roughly square shape, which is indicative of a well defined coercive voltage, i.e. most of the dipoles have been switched at the coercive voltage level of applied electric field. As in FIG. 1A, two stable states 12B and 14B are shown in FIG. 1B, which represent opposite alignments of the internal dipoles after the applied electric field has been removed.
To be useful, a ferroelectric material in a memory array must have the ability to retain :data when repeatedly switched between the two stable data states 12A and 14A as shown in FIG. 1A, or data states 12B and 14B as shown in FIG. 1B. Conventional non-volatile memory cells such as EEPROMs are non-destructive readout devices in that reading the data state does not change that data state. Conventional non-volatile memories allow an unlimited number of read cycles, but a very limited number of write cycles. Ferroelectric memory devices are typically destructive readout devices in that switching occurs for each reading of the data. Consequently, it is desirable that the number of cycles before failure of the ferroelectric memory cell be very much higher than that for EEPROMs or on the order of magnitude of one billion or one trillion cycles.
It is also desirable in many applications for the ferroelectric memory cell to operate at low voltage such as 3.3 volts. Thus, the coercive voltage should be significantly lower than the desired operating voltage of the memory. The minimum voltage at which the switched charge reaches its maximum value, the saturation voltage, should also be about the same as or lower than the desired operating voltage for optimum performance. It is further desirable that the ferroelectric material in a memory array retain a data state over prolonged periods of time at elevated temperatures. Ideally, retention of a data state should be independent of the state of prior data storage. The bismuth layered ferroelectric compounds referred to above are thus ideally suited for a ferroelectric memory application because of their desirable cycling, low voltage, and retention properties.
While ferroelectric bismuth compounds are well suited to operation in a memory because of their electric characteristics, they are difficult to manufacture as a capacitor dielectric material in a typical semiconductor processing flow. A classical method of depositing bismuth titanate, or other bismuth ferroelectric materials, is to sputter the material from a single target at elevated temperatures onto crystalline planes of a substrate oxide. The substrate oxide is chosen so that the lattice dimensions are matched to the lattice dimensions of the ferroelectric bismuth compound, which results in a crystalline film that is ferroelectric. The classical sputter deposition method suffers from several shortcomings.
The requirement of lattice matching severely restricts the possible choices of substrates. The common substrates used to lattice match with the ferroelectric film, such as aluminum oxide, lanthanum aluminum oxide, magnesium aluminum oxide or magnesium oxide, are not conductive. In order to fabricate a capacitor, conductive electrodes are required in order to apply an electric field across the ferroelectric film. To use these films on insulating lattice matched oxide substrates, the ferroelectric film must be physically peeled from the substrate and electroded. After separation from the substrate, the films are very fragile and difficult to handle. Therefore the process of using such films is very difficult and unsuitable for fabrication of conventional nonvolatile memories. It is conceivable that there might exist a conductive oxide that could also provide lattice matching, but the known conductive oxides suffer from poor conductivity, thermal instabilities, and complex fabrication methods.
Another shortcoming of the classical method is the requirement of elevated deposition temperatures, e.g. 500.degree. to 700.degree. C. These very high temperatures add complexity and impose severe constraints on the deposition equipment. The high temperatures also can degrade the bottom electrode surface flatness due to hillock formation.
Alternative prior art spin-on deposition methods, such as sol gel or metal organic deposition, may yield switching bismuth layered oxide films on conductive electrodes, but these methods also suffer from several shortcomings. The orientation of grains and grain size of the deposited film might not be suitable or controllable when used on bottom electrodes suitable for conventional integrated circuit processing, resulting, in a low grade ferroelectric material that yields little switched charge. Furthermore, these spin-on deposition methods suffer from practical incompatibilities with the processing techniques used in the fabrication of conventional semiconductor integrated circuit non-volatile memories. Spin-on deposition planarizes the surface topography on top of an integrated circuit by becoming thicker in the troughs and thinner at elevated surface features. This non-uniform film thickness deposition causes problems during subsequent patterning steps, resulting in regions of under etching and regions of over etching. Another problem with spin-on deposition techniques is that the availability of suitable precursor solutions is not as commercially developed as the availability of sputter targets for the deposition process referred to above. The purity requirement of near zero mobile ion content is readily achieved with commercially available metal oxide powders used for sputtering targets, but remains a challenge for spin-on sources. The spin-on process is susceptible to a host of additional problems such as precipitation of detrimental particles in the solution or during application, streaking, poor adhesion, and film cracking due to volumetric changes during solvent evolution bakes.
Another alternative prior art deposition method is the technique of "co-deposition" in which two or more targets are used, primarily to control the composition of a deposited film. Typically, the deposition from multiple targets is either done simultaneously or separately, but in the latter case with extremely thin layers. The layer thickness may be only a few Angstroms thick, in which case there is an effective mixing of layers to form a homogeneous film composition. This technique, as now commonly practiced, is not useful for forming bismuth layered ferroelectric films. While the resultant film composition is made of the desired constituent elements, and in the proportions required, the film is not ferroelectric. It is speculated that initial orientation of the crystal grains, or perhaps other properties of the film, result in a polarization axis that is not perpendicular to the plane of the metal electrodes in contact with the film. The capacitors made with such a film, therefor, are not ferroelectric in that a significant switched charge as can be detected with a conventional sense amplifier is not produced in response to an applied electric field.
A need remains, therefore, for a method for the formation of desirable bismuth ferroelectric layers with consistent desirable electrical properties that is readily integrated and compatible with conventional complementary metal oxide semiconductor transistor ("CMOS") processing.